The emergence of fifth-generation mobile communication
networks (5G) is driving the adoption of the Internet of
Things (IoT) for various industrial, medical, and mili-
tary applications. However, the reliability and security of
data is fast becoming a key bottleneck for further IoT
growth. Fig. I(a) shows an overview of a typical smart
home/building system, where multiple IoT devices are con-
nected on trusted gateways, and initially-processed private
information is uploaded into a cloud database for further data
analytics and decision making. While the wireless communi-
cation links can be encrypted for security, each device’s data
is still vulnerable to physical side-channel attacks on the node
itself (e.g., using near-field probes or power supply moni-
tors). Fundamentally, this is due to the fact that encryption is
generally not energy-efficient enough for use in short-range
The resulting security vulnerabilities are unacceptable for
IoT devices in critical systems such as aircraft, military
facilities, and hospitals. These devices thus require effective
mechanisms to ensure the confidentiality, authenticity, and
purity of board- or chip-level wired data links. However,
these requirements pose challenges for traditional private- or
public-key cryptographic schemes, since IoT devices usually
have stringent requirements on memory, cost, and processing
power. The security of alternative encryption schemes suit-
able for recourse-constrained platforms highly depends on
the effective volume of keyspace. Traditionally, static entropy
is simply generated off-chip as security keys that are stored
in a nonvolatile manner. Such key storage is well known to be
vulnerable to a wide range of software and physical attacks,
such as noninvasive probing. Some
other high-fidelity encryption methods (such as e-wallets
for cryptocurrencies) provide up to 192 bit keyspace but at the cost of high power, memory, and increasing circuit
complexity, making them hard to implement on practical IoT
devices. Physically unclonable functions (PUFs) have been
extensively explored for generating static on-chip encryp-
tion keys using chip-specific random variation [1]. Typically,
weak PUFs suffer stability issues while strong PUFs have a
low density of bits per unit area. Also, PUFs are not popu-
lar for communications due to i) the need for an expensive
enrollment (key sharing) phase before in-field operations [1],
and ii) the fact that they are non-configurable due to using
a static entropy generation mechanism. Dynamic entropy
generation by random number generators (RNGs) [2], [3]
provides another fundamental primitive to generate on-the-
fly random bits (only available during power-on condition)
for the creation of fresh session keys. While RNG-based
digital encryption does provide sufficient data security, it has
significant on-chip power and data overhead.
Recently, chaos-based encryption has emerged as an attrac-
tive alternative to the methods mentioned above due to
the highly unpredictable and apparently random nature of
chaotic signals [4], [5]. Chaotic systems are extremely sen-
sitive to changes in parameter values and initial condi-
tions, which can be used to produce a large number of
uncorrelated chaotic sequences for secure communications
[6]–[9]. A comprehensive survey of wireless chaos-based
communications systems can be found in [7]. In chaos-based
communication systems with coherent detection, chaotic sig-
nals are required to synchronized at both transmitter and
receiver side in order to recover the transmitted data. Thus,
unlike the above-mentioned RNG-based digital encryption
methods, here we utilize the extreme parameter sensitivity
of chaotic systems as private encryption keys for data com-
munications. This approach also results in a simple enroll-
ment process: security keys are shared simply by using a pair of chaotic systems with the same set of parameters,
in which case generalized synchronization methods [10], [11]
can be applied to guarantee successful data decryption. The
effects of unavoidable parameter mismatches can also be
minimized using recently-proposed adaptive synchronization
methods [4], thus improving robustness. One reported draw-
back of this approach is that low-dimensional chaotic systems
are potentially broken using known-plaintext attacks [12].
Thus, high-dimensional and hyperchaotic systems are desir-
able for encryption